Back in February 2017, Valve revealed that it was implementing a new approach to catch out cheaters in Counter-Strike: Global Offensive. You’ll often find users online complaining about cheaters in CS:GO, and VAC bans usually come in slow waves. To tackle this, Valve turned to machine learning to catch them out, and now we know a bit more about the system.
VACnet is a new server farm running machine learning techniques to analyse CS:GO matches and detect cheats. Valve programmer, John McDonald talked about the system at GDC last week. Research in to VACnet began in 2016, when complaints of cheating in CS:GO reached an all-time high. Rather than continuing the arms race where Valve detects/bans cheats and developers create new ones, the team decided to implement machine learning to adapt over time and hopefully catch people out faster.
Currently, VACnet trains itself from monitoring matches and submitting cases where a cheat has been flagged to Valve. The CS:GO team then reviews the cases submitted and will determine guilt, which in turn, helps the algorithm learn how to properly detect cheats. Having humans review each case will also help stop users from being burned by false bans as VACnet gets off the ground.
As PCGamer reports, the VACnet servers are powered by 54 CPU cores and 128GB of RAM per blade. There are 16 blades per chassis, and four chassis in total. In all, that means VACnet is powered by 3,456 processors in total. Currently, Valve only needs 1,700 CPUs to handle all of the data coming in from CS:GO each day, but it decided to throw twice as much power at the system to leave room for future expansion.
Discuss on our Facebook page, HERE.
KitGuru Says: CS:GO does have its share of cheaters but I think the issue can be blown out of proportion at times. Without a killcam system in competitive mode, it can be easy to call foul play if you don’t get to see how things played out from the other player’s perspective. Still, it sounds like VACnet is a good step forward, and will hopefully result in fairer competitive matches going forward.